[Top] [All Lists]

Re: Asynchronous crypto layer.

To: Michal Ludvig <michal@xxxxxxxx>
Subject: Re: Asynchronous crypto layer.
From: Evgeniy Polyakov <johnpol@xxxxxxxxxxx>
Date: Fri, 29 Oct 2004 18:36:06 +0400
Cc: netdev@xxxxxxxxxxx, cryptoapi@xxxxxxxxxxxxxx, hadi@xxxxxxxxxx
In-reply-to: <41824D9A.3070407@xxxxxxxx>
Organization: MIPT
References: <1099030958.4944.148.camel@uganda> <1099053738.1024.104.camel@xxxxxxxxxxxxxxxx> <20041029180652.113f0f6e@xxxxxxxxxxxxxxxxxxxx> <41824D9A.3070407@xxxxxxxx>
Reply-to: johnpol@xxxxxxxxxxx
Sender: netdev-bounce@xxxxxxxxxxx
On Fri, 29 Oct 2004 16:03:06 +0200
Michal Ludvig <michal@xxxxxxxx> wrote:

> Evgeniy Polyakov told me that:
> > On 29 Oct 2004 08:42:18 -0400
> > jamal <hadi@xxxxxxxxxx> wrote:
> > 
> >>>I am very curious if you see perfomance improvements over old scheme
> >>in the case of a single crypto chip in a fast CPU.
> >>It has been shown in the past that with a xeon in the range of 2Ghz the
> >>context setup and the big lock (and lack of async) implied that
> >>performance enhancement using a crypto chip was negligible.
> >>IIRC, the only time it started showing anything useful was when a
> >>compute intensive alg like 3DES was chewing packets >= 1000 bytes.
> >>
> >>My suspicion is you will show it is better to use a crypto chip with
> >>async;-> In the minimal you should show some improvement.
> > 
> > 
> > If we have a hardware accelerator chip, than we _already_ have improvements 
> > with even the worst async crypto layer, since software and hardware 
> > will work in parrallel.
> > I agree that multigigahertz box will beat my HIFN card, 
> > but I doubt it can beat 1gghz VIA.
> I have a very preliminary driver for FastCrypt PCI board for 3DES at
> For now it works with some very ugly hacks in the current cryptoapi, but
> I can give it a try with your acrypto and report the results.

It would be very appreciated>
> I admit I haven't read your sources too deeply yet so excuse me a dumb
> question - does acrypto replace or extend cryptoapi? Once I get it
> running will it take over e.g. encryption for IPsec?

They are both an addendum to each other, 
any crypto layer can be "tunrned into compatibility mode" - 
i.e. anyone can write bridges like attached sha1_provider.c (it is bridge
from async into sync mode). So I can not answer - but in _current_ 
implementation without any kind of bridges it is an extension.

If you compile this sources then you will still have old sync behaviour, 
IPsec and any other old-style application should be rewritten(like 
attached consumer.c) to use new asynchronous crypto layer features.
> Michal Ludvig
> -- 
> * A mouse is a device used to point at the xterm you want to type in.
> * Personal homepage -
> _______________________________________________
> Subscription:
> List archive:

        Evgeniy Polyakov

Only failure makes us experts. -- Theo de Raadt

<Prev in Thread] Current Thread [Next in Thread>